NEWS

zipfR 0.6-70 (2020-10-11)

parallelization of lnre.bootstrap(), using forked processed on Unix-like platforms or with a properly initialized snow cluster (based on the 'parallel' package)
bootstrapped confidence intervals in lnre() are speeded up substantially
additional productivity measures and some other improvements to productivity.measures()
lnre.productivity.measures() computes approximate expectations for samples from LNRE population or estimates sampling distribution by parametric bootstrapping
plot.tfl() now also plots relative frequencies (pmw) for comparison across different sample sizes, as well as population distributions described by LNRE models
plot method (plot.lnre) for LNRE models, showing either the log-transformed type density function or the distribution function for a quick visualisation
all plot methods accept list of objects in first argument to avoid do.call() trickery, but need to be called explicitly rather than via method dispatch in this case
lnre() now allows a user-defined cost function in parameter estimation (e.g to fit against Zipf ranking = top part of TFL, or to set parameters based on implicit constraints)
bug fix: tfl2spc() no longer includes types with f = 0 in frequency spectrum, and spc() now complains in such cases
new default colour scheme and line styles

default minimization algorithm changed to "Nelder-Mead", since "Custom" often failed to converge on problematic data sets
new default cost function "gof" based on the multivariate chi-squared statistic, which is closely related to maximum likelihood estimation
parameter optimization now performs multiple runs with different random start values by default, improving robustness of the parameter estimation
lnre.bootstrap() implements parametric bootstrapping from an LNRE model, and the function bootstrap.confint() provides robust estimates of confidence limits
rlnre() can directly generate type-frequency lists instead of token vectors, which is much more efficient for large samples (and essential for large-scale bootstrapping experiments)
bootstrapped confidence intervals for model parameters can be obtained with confint() applied to an LNRE model object with bootstrap replicates
compute various productivity measures from observed frequency spectrum, type-frequency list, vocabulary growth curve or token vector
merge() method for two or more type frequency lists
new example data sets from Evert & Luedeling (2001) and Baayen (2001)

maintenance release for compatibility with new CRAN checks and restrictions
zipfR plotting utilities (zipfR.begin.plot, zipfR.end.plot, etc.) are deprecated; device="x11" and device="quartz" now both open the default device with dev.new(), which is not guaranteed to be an on-screen device
update stale URLs and maintainer e-mail addresses

zipfR tutorial has been rewritten as genuine package vignette
write.tfl() and read.tfl() now allow character encoding of the disk file to be declared; read.tfl() safely reads type strings containing quotes

interim release for compliance with new, stricter CRAN requirements
Zipf-ranking plots for type-frequency lists (of class "tfl")
file I/O automatically compresses/decompresses ".gz" files in functions read.tfl(), read.spc(), read.vgc(), write.tfl(), write.spc(), write.vgc()
dropped support for very old R versions (before 2.3.1 or so), where read.delim() does not accept comment.char= option; now requires 2.10.1+
removed zipfR.legend() function since standard legend() now offers the same convenient placement options
added citation details (file inst/CITATION), which can be displayed with citation(package="zipfR")
upgraded to GPL v3

various bug fixes and minor improvements
improved robustness and tolerance for rounding errors in expected frequency spectra

first public release on CRAN
improved parameter estimation, can be fine-tuned by users with choice of cost functions and minimization algorithms
default settings work well for most data sets
minor bug fixes and improved documentation
updated version of tutorial